Future proofing cities

a 10 day project with the Le Wagon Data Science & Machine Learning bootcamp

Description

This repository is readily available for anyone wanting to evaluate heat distribution in an urban landscape. The aim of the project is to use a convolutional neural network to predict heat islands within a city, which we defined as areas of relatively high land surface temperature (LST). This Neural Network takes 15 features determining a city's topography, land cover types, building height and density and will predict the difference to the mean temperature for each pixel. Pixels are defined as 70*70m and are fed into the Network as a tensorflow object of shape (1, 15)

Features:

• Average Building Density/pixel
• Average Building Height/pixel
• Elevation
• Landuse Type (12 categories)

Target:

• Difference to the Mean of Land Surface Temperature/ pixel

Installation & usage

Data

• Preprocessed Data for Land Cover for Paris and Berlin: data/processed_data/Berlin/Berlin_landuse.csv

• Preprocessed Data for Elevation: data/processed_data/Berlin/Berlin.csv

• Preprocessed Data for Building Height and Density ohsome API on building data

Model modules/ml_logic/model.py

Training the model

• Install dependencies

make install

• Train on Paris data

make run_train

Predicting using docker

• Copy content from .env-sample to .env and update your relevant information
• Run

pyenv allow

• Build docker image

docker build -t $IMAGE:prod .

• Run docker image on port 8000

docker run -it -e PORT=8000 -p 8000:8000 --env-file .env $IMAGE:prod

• Access the docs on localhost:8000/docs

Streamlit frontend

Our Git Repository which contains the information for the deployment of our user interface:

https://github.com/b-fa-ce/future-proofing-cities-frontend

Project Contributors:

• Bruno Faigle-Cedzich
• Matt Hall
• Afanasis Kiurdzhyiev
• Leah Rothschild

Project created and developed in the context of finalising our Data Science & Machine Learning Course with Le Wagon